Objective: To develop and validate a risk-classification system for in-hospital death, clinically useful for general hospital adult primarily non-surgical cases.
Methods: Admissions for non-surgical conditions at 5 public general hospitals of Minas Gerais were included. Procedures: Build a predictive model for death during admission, using logistic regression; Create a severity index based on the independent effect of the selected variables, and then, validate its ability to predict in-hospital death during index admission; Validate the predictive scale by challenging it with a new dataset.
Results: The final multivariate model included seven significant predictive variables: age, gender, diagnostic-related group, hospital of index admission, admission to the ICU, total length of stay, and unplanned surgical procedure. This model presented adequate fit and fair discriminative performance (AUC=0.78). Temporal validation with a new sample also presented an adequate fit, and the discriminative performance was again fair (AUC=0.76).
Conclusions: A dynamic and clinically useful risk classification system for in-hospital death of non-surgical admissions has been validated.